Abstract

Traditional geographic data statistical modeling focuses on global spatial autocorrelation and time dynamic effects, but ignores the multi-scale structure, scale effects and local spatial abruptness of geographic data, leading to biased model parameter estimation and invalid statistical inference.To solve this problem, we first developed an adaptive spatial multi-scale statistical modeling method and a Bayesian Markov chain Monte Carlo simulation algorithm for model estimation.At the same time, the global spatial dependence effect, local variation effect and scale effect of geographic data were modeled to improve the accuracy and effectiveness of parameter estimation.In addition, for the complex time-space structure of geographic big data such as GPS trajectories, we developed a multi-scale time-space dynamic statistical model and its estimation algorithm. These multi-scale temporal spatial statistical models have been applied in the fields of geography and health, quality of life and urban economics.

Presenter Profile

Dong Guanpeng graduated from the University of Bristol with a Ph.D. He is currently a lecturer at the School of Geography and Planning, University of Liverpool, and a Visiting Fellow at the Institute of Quantitative Methods, University of Sheffield. In 2016, he received the Jon Rasbash Prize in the Quantitative and Computational Social Sciences from the Council of Economic and Social Sciences (ESRC).He has published more than 20 SSCI/SCI papers on spatial analysis and GIS in internationally renowned journals, and developed the HSAR open source statistical software package.